LTE Superband: What you need to know about (1)

In this series of posts we will be diving into the Huawei LTE Superband features. In this 1st post we will understand the Multi-carrier Unified Scheduling feature and the advantages of implementing it.

LTE Superband: Multi-Carrier Unified Scheduling Feature

This set of features aims to optimize the resource efficiency of single large-bandwidth carriers. In multi-carrier scenarios, it is very likely to find non-carrier aggregation (CA) UEs or CA UEs that cannot aggregate as many carriers as configured by the network.

In this scenarios, multi-carrier scheduling includes inter-carrier UE scheduling (or carrier selection) and inter-carrier data scheduling. Multi-carrier scheduling differs from single-carrier large-bandwidth scheduling in the following aspects:

UEs do not report the spectral efficiency of neighboring E-UTRA frequencies. Hence, eNB cannot accurately select optimal carriers for the UEs.
Inter-carrier data scheduling requires one or more procedures of secondary serving cell (SCell) configuration, activation, and data split. This process can take several seconds. Furthermore, data distribution from the primary serving cell (PCell) to an SCell requires multiple information exchanges, increasing the latency. This latency prevents UEs from using large bandwidths continuously for transmission causing the radio resource efficiency lowers.
Since coverage differs among frequency bands, the eNodeB restricts the maximum coverage of each carrier by imposing a signal level threshold and using other measures. This reduces the overlapping coverage area of multiple carriers. This causes the UE at the coverage edge of a low-frequency cell, to unlikely benefit from CA or to use the optimal carrier or carrier combination among multiple candidates.

Huawei LTE Superband Multi-Carrier Unified Scheduling features uses following functionalities to improve resource efficiency of single carriers.

Virtual grid technology

By using the multi-dimensional measurements of radio signals, UEs with the same radio characteristics are categorized as a single group. For example, if the measurement result on a frequency reported by UE 1 is [(Cell 1, RSRP 1), (Cell 2, RSRP 2), …] and that reported by UE 2 is the same, the eNodeB considers the two UEs to be in the same virtual grid.
Then by using machine learning technology on the virtual grids data collected from the measurements. The main goal is to explore the mapping in signal characteristics from all virtual grids in a cell to a frequency.
Virtual grid models include RSRP prediction models and spectral efficiency prediction models. After obtaining a UE’s virtual grid information on a frequency, the eNodeB can query the prediction model for each neighboring E-UTRA frequency. Then it can quickly estimate the RSRP of the UE on each neighboring frequency. Spectral efficiency prediction model development also uses this process.

Before get into the features of LTE Superband Multi-Carrier Unified Scheduling, let’s dive on how the virtual grid works.

The virtual grid model building process is as follows:

The eNodeB determines the scope of cells and neighboring E-UTRA frequencies for which virtual grid models are to be built. The eNodeB does not build virtual grid models for the following cells:
- Cells using MultiRRU.
- Cells configured as high speed mobility.
- eMTC only cells.
- FDD cells with radius above 100Km.
- FDD cells with NB-IoT
- Cells configured as DL only or LAA

The eNodeB does not build virtual grid models for the following neighboring E-UTRA frequencies:

Cells in Frequency band 46
Cells with configuration forbidding virtual grid model build.

If the number of virtual grid models to build exceeds the limit set by hardware and CA feature, the eNodeB calculates the priorities of these candidate models and then selects higher-priority candidate models to build as long as the limit allows. The priorities of the models are calculated based on the following sequence of factors:

Model can be Intra-eNodeB/Inter-eNodeB, depending on the neighboring E-UTRA frequency considered for the calculation. An intra-eNodeB model has a higher priority than an inter-eNodeB model.
If a model is for a larger-bandwidth cell, the model has a higher priority.
The camping priority.
If a model is for a larger-bandwidth neighboring E-UTRA frequency, the model has a higher priority. This factor applies only to intra-eNodeB models.

2. The eNodeB randomly selects 15 UEs in individual cells constantly and collects their measurement reports on different frequencies as sample data for RSRP prediction model training. In addition, the eNodeB collects the PRB usage of the neighboring cells of the strongest cell on each frequency and the UEs’ reception capabilities and spectral efficiency on each frequency as sample data for spectral efficiency prediction model training. The eNodeB collects 24-hour sample data. If the data is too few for building a model, the eNodeB waits for around 1 month before it tries to build the model again.

3. When data collection is complete, the eNodeB starts the model training process and calculates the accuracy of each model.

4. The eNodeB checks the model accuracy to evaluate whether it reaches conditions for launching models. If the accuracy of an RSRP prediction model is greater than or equal to certain threshold, the eNodeB acknoledge reaching the condition so the model has been built successfully, and it puts the model into service. Otherwise, the eNodeB considers that the model has failed to be built. If the building process fails several consecutive times, the eNodeB waits for around 1 month before it tries to build the model again.

5. After a model created on a 24-hour sample data is put into service, the eNodeB randomly selects three UEs in the cell around every minute. It collects sample data constantly to build a new model. If the new model is more accurate than the old model, the new model replaces the old one.

The eNodeB monitors the performance of virtual grid models and updates them as follows:

After a model is put into service, the eNodeB starts the model monitoring and update procedure.
It monitors the Handover success rate in real time while the model is active. If the KPI are lower than expected, the eNB suspends the use of the model and the eNB consider the model is out of date, so it suspends the model.
Every third day, the eNB perform model accuracy test. Same as previous step, if the model accuracy is below threshold, the eNB consider the model as out of date and then suspends the model.
It is important to notice that the eNB randomly selects 3 UEs in the cell every 40s and collect sample data to constantly rebuild the model.

In general, virtual grids comes from A3 measurement events. The virtual grid models of the serving cell may not provide the spectral efficiency of some frequencies. In this case, inter-frequency event A4/A5 measurement results of the UE can be used to determine the virtual grid where the UE is located on a neighboring E-UTRA frequency. Then, the eNodeB uses the spectral efficiency prediction models of the strongest cell on that frequency.

For LTE Superband Multi-Carrier Unified Scheduling set of features, the virtual grid technology is the basic functionality for LTE Multi-Carrier Unified scheduling set of features. Now, let’s review the 3 Huawei features using this techonology and how they can benefit the performance of the network.

1. LTE Superband: Smart carrier selection

Summary: When UEs are accessing the network, the eNodeB selects carriers that can provide the maximum downlink throughput for the UEs. It basically consideres the bandwidth, load, and spectral efficiency predicted based on virtual grids. Periodic evaluation is performed for UEs that stay online for a long time. This way the UEs are always using the optimal carriers or carrier combinations. This process allows more UEs reahcing higher spectral efficiency without the need for measurement gaps.

This LTE superband feature allows the eNB to calculate the priority of each candidate carrier or carrier combination, It then assigns the highest-priority candidate to a UE. The calculation is based on:

Downlink air interface capability: Air interface capability is the primary factor in the calculation of the priorities of candidates. The eNodeB assigns the carrier or carrier combination with the highest downlink air interface capability to the UE to increase the downlink user-perceived throughput. This capability evaluation considers the UE-level downlink spectral efficiency, cell bandwidth, and cell load:
- UE-level downlink spectral efficiency: Using the virtual grid models, the eNB can inquire the UE-level downlink spectral efficiency, however this only works when the predictions cover all frequencies.
- Downlink cell bandwidth: It considers the bandwidth available and the cell capacity scale factor used in MLB.
- Cell load: Depending whether the cells supports Massive MIMO, carrier agregation etc. the calculcation for cell load varies.
Downlink bandwidth: Candidate with larger PRB capacity has a higher priority
PCC priority
Handover, if handover is necessary after a candidate is selected for a UE, then the priority is lower because, haondvers interrupts data transmission for short time.
SCell change causes a transient decrease in the data rate, so if it is necessary then it decreases the probability.
SCC priority
Number of carriers If two candidates have the same aggregated bandwidth, the candidate that contains fewer carriers has a higher priority. This is because fewer carriers consume fewer PDCCH/PUCCH resources.

The basic procedure is as follows:

LTE superband. Process for smart carrier selection

In general, after the virtual grid model is built and when UE is accessing the network, the eNB can use the model to predict the best carrier for the UE based on its latest measurements. This way the UE can be transferred (by Handover of configuring SCell). So, this LTE superband function enables the eNodeB to select optimal carriers for UEs, this is based on factors such as the UE-level spectral efficiency and load. Therefore, the average downlink UE data rates increase.

Take into consideration that, even though this LTE superband feature provides a good way to efficiently use the spectrum available, it does bring some impacts:

The UE distribution across frequency bands changes. Therefore, band-specific performance indicators vary.
Virtual grid modeling has a direct impact on CPU usage.
The number of handovers increases because the eNodeB needs to select better carriers for UEs
There are more signaling messages over the air interface, and the signaling radio bearer (SRB) traffic increases. This can impact delivery of RRC connection reconfiguration messages The average MCS index may decrease because SRBs use MCSs with small indexes. In addition, RRC connection reconfiguration messages, increasing the service drop rate.
For fast-moving UEs that cannot be identified, the selected carriers or carrier combinations may not be optimal throughout the service duration causing impact on user throughput.

2. LTE Superband: Ultra-low-latency scheduling

Summary: This functionality ensures that SCells are configured and activated in an ultra-fast manner. The virtual grid technology helps SCells to be configured much faster than when measurements are required. Delayed SCell deactivation reduces the number of times SCells are deactivated and activated, mitigating the performance loss caused by SCell activation latency. Ultra-low-latency interaction between the PCell and SCells allows for simultaneous scheduling in the PCell and activated SCells.

This feature basically works implementing following functionalities:

Delayed SCell deactivation: When the downlink buffer for a UE that has an SCell activated is cleared, an SCell deactivation waiting timer starts. If the buffer keeps empty before the timer expires, the eNodeB deactivates the SCell for the UE. If the configuration is correct, SCells can remain in the active state for most UEs that have frequent bursts of data transmission. The eNodeB no longer needs to wait for SCell re-activation and can earlier receive valid channel state information (CSI) related to SCells from the UEs. Therefore, the data transmission performance of SCells improves.
Accurate data split with ultra-low latency
- In the intra-eNodeB CA and inter-eNodeB CA based on eNodeB coordination scenarios, data distribution to SCells and PUCCH resource allocation start as early as in the transmission time interval (TTI) when the UE buffer changes to be not empty. In this way, simultaneous scheduling in the PCell and SCells is achieved. This function takes effect in the inter-eNodeB CA based on eNodeB coordination scenario only when this function is enabled on all the eNodeBs involved.
- The volume of data distributed to each CC depends on the bandwidth and spectral efficiency of each CC and the PCC load. This facilitates accurate data split and reduces resource waste.

This function increases the SCell usage, increasing the average downlink data rate of UEs, especially in scenarios with frequent bursts

Some of the impacts this LTE superband feature has are:

The UE distribution across frequency bands changes. Therefore, band-specific performance indicators vary. This also applys for the distribution beween CA UEs and non-CA UEs.
The CPU usage increases slightly.
Increase of overhead of UL and DL control channel for SCells.
RBLER can increase impacting retainability.

3. LTE Superband: Coverage threshold adaptation

Summary: Once the virtual grid technology is working, the coverage thresholds adapts for multiple process such as: SCell configuration/Removal, coverage-based inter-frequency handovers. This enlarges the overlapping coverage area between multiple frequency bands.

Coverage threshold adaption enlarges the overlapping coverage area of multiple bands. Basically, by modifying the thresholds related to CA, Handover etc. the coverage can be extended.

LTE superband coverage threshold adaptation has the following impacts on the network:

The UE distribution across frequency bands changes. Therefore, band-specific performance indicators vary.
The handover success rate decreases and the service drop rate rises due to the extended cell edge.
Neighbor interference increases.